Self-driving production: Autonomous agents for incident response

At HumanX 2026, Michael Grinich interviewed Anish Agarwal, CEO of Traversal, about the company's innovative approach to automating incident response in production systems using autonomous agents. Traversal's solution involves a "production world model," a comprehensive representation of every component and dependency within a system, enabling autonomous troubleshooting much like self-driving cars navigate using world models. The company addresses the challenge of processing vast amounts of telemetry data by employing an AI-native compressor to reduce data volume without losing essential signals. Traversal applies an autonomy framework similar to the SAE levels for self-driving cars, focusing on the transition from manual to fully autonomous operations. A significant hurdle in reaching full autonomy lies in organizational change rather than technology, as engineers are eager to design resilient systems rather than constantly manage incidents. The ultimate challenge is achieving causal reasoning in AI, a task that goes beyond correlation and requires understanding cause and effect, which could have far-reaching applications beyond just site reliability engineering. Traversal aims to fill the gap in AI investment for production systems, which remains a bottleneck despite advancements in AI-driven software development processes.